The invention relates to the technical field of rock
mass stress field inversion, in particular to a stratified rock
mass discrete element model crustal stress field inversion method based on
deep learning, which comprises the following steps: acquiring a main control surface patch node extraction normal change sequence, comparing attitude path positioning direction
mutation regions, analyzing included angle change to generate a construction path trend structural body, and calculating the
stress field of the stratified rock
mass. And node information is extracted to establish a structural path combination input primitive set, direction attributes are written to form a crack direction embedded tag group, and a constructed inversion sequence is input into a preset
deep learning model to output a
crustal stress inversion
tensor field result. According to the method, the node direction change is extracted,
mutation fragments are identified, spatial paths consistent in direction are constructed, a node sequence is generated in combination with the
principal stress direction and the construction trend, coordinates, direction and construction information are fused to construct a spatial
label group, an inversion input sequence is embedded, and the structural path and
stress distribution relation is associated. The space matching between the path identification capability and the tensile stress direction is improved, and the distribution continuity and inversion stability of the
tensor result in the structure are enhanced.